2 * H.26L/H.264/AVC/JVT/14496-10/... encoder/decoder
3 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
5 * This file is part of FFmpeg.
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
23 * @file libavcodec/h264.h
24 * H.264 / AVC / MPEG4 part10 codec.
25 * @author Michael Niedermayer <michaelni@gmx.at>
28 #ifndef AVCODEC_H264_H
29 #define AVCODEC_H264_H
31 #include "libavutil/intreadwrite.h"
34 #include "mpegvideo.h"
36 #include "rectangle.h"
38 #define interlaced_dct interlaced_dct_is_a_bad_name
39 #define mb_intra mb_intra_is_not_initialized_see_mb_type
41 #define LUMA_DC_BLOCK_INDEX 25
42 #define CHROMA_DC_BLOCK_INDEX 26
44 #define CHROMA_DC_COEFF_TOKEN_VLC_BITS 8
45 #define COEFF_TOKEN_VLC_BITS 8
46 #define TOTAL_ZEROS_VLC_BITS 9
47 #define CHROMA_DC_TOTAL_ZEROS_VLC_BITS 3
48 #define RUN_VLC_BITS 3
49 #define RUN7_VLC_BITS 6
51 #define MAX_SPS_COUNT 32
52 #define MAX_PPS_COUNT 256
54 #define MAX_MMCO_COUNT 66
56 #define MAX_DELAYED_PIC_COUNT 16
58 /* Compiling in interlaced support reduces the speed
59 * of progressive decoding by about 2%. */
60 #define ALLOW_INTERLACE
62 #define ALLOW_NOCHROMA
65 * The maximum number of slices supported by the decoder.
66 * must be a power of 2
70 #ifdef ALLOW_INTERLACE
71 #define MB_MBAFF h->mb_mbaff
72 #define MB_FIELD h->mb_field_decoding_flag
73 #define FRAME_MBAFF h->mb_aff_frame
74 #define FIELD_PICTURE (s->picture_structure != PICT_FRAME)
79 #define FIELD_PICTURE 0
81 #define IS_INTERLACED(mb_type) 0
83 #define FIELD_OR_MBAFF_PICTURE (FRAME_MBAFF || FIELD_PICTURE)
86 #define CHROMA h->sps.chroma_format_idc
92 #define CABAC h->pps.cabac
95 #define EXTENDED_SAR 255
97 #define MB_TYPE_REF0 MB_TYPE_ACPRED //dirty but it fits in 16 bit
98 #define MB_TYPE_8x8DCT 0x01000000
99 #define IS_REF0(a) ((a) & MB_TYPE_REF0)
100 #define IS_8x8DCT(a) ((a) & MB_TYPE_8x8DCT)
103 * Value of Picture.reference when Picture is not a reference picture, but
104 * is held for delayed output.
106 #define DELAYED_PIC_REF 4
124 NAL_AUXILIARY_SLICE=19
131 SEI_BUFFERING_PERIOD = 0, ///< buffering period (H.264, D.1.1)
132 SEI_TYPE_PIC_TIMING = 1, ///< picture timing
133 SEI_TYPE_USER_DATA_UNREGISTERED = 5, ///< unregistered user data
134 SEI_TYPE_RECOVERY_POINT = 6 ///< recovery point (frame # to decoder sync)
138 * pic_struct in picture timing SEI message
141 SEI_PIC_STRUCT_FRAME = 0, ///< 0: %frame
142 SEI_PIC_STRUCT_TOP_FIELD = 1, ///< 1: top field
143 SEI_PIC_STRUCT_BOTTOM_FIELD = 2, ///< 2: bottom field
144 SEI_PIC_STRUCT_TOP_BOTTOM = 3, ///< 3: top field, bottom field, in that order
145 SEI_PIC_STRUCT_BOTTOM_TOP = 4, ///< 4: bottom field, top field, in that order
146 SEI_PIC_STRUCT_TOP_BOTTOM_TOP = 5, ///< 5: top field, bottom field, top field repeated, in that order
147 SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM = 6, ///< 6: bottom field, top field, bottom field repeated, in that order
148 SEI_PIC_STRUCT_FRAME_DOUBLING = 7, ///< 7: %frame doubling
149 SEI_PIC_STRUCT_FRAME_TRIPLING = 8 ///< 8: %frame tripling
153 * Sequence parameter set
159 int chroma_format_idc;
160 int transform_bypass; ///< qpprime_y_zero_transform_bypass_flag
161 int log2_max_frame_num; ///< log2_max_frame_num_minus4 + 4
162 int poc_type; ///< pic_order_cnt_type
163 int log2_max_poc_lsb; ///< log2_max_pic_order_cnt_lsb_minus4
164 int delta_pic_order_always_zero_flag;
165 int offset_for_non_ref_pic;
166 int offset_for_top_to_bottom_field;
167 int poc_cycle_length; ///< num_ref_frames_in_pic_order_cnt_cycle
168 int ref_frame_count; ///< num_ref_frames
169 int gaps_in_frame_num_allowed_flag;
170 int mb_width; ///< pic_width_in_mbs_minus1 + 1
171 int mb_height; ///< pic_height_in_map_units_minus1 + 1
172 int frame_mbs_only_flag;
173 int mb_aff; ///<mb_adaptive_frame_field_flag
174 int direct_8x8_inference_flag;
175 int crop; ///< frame_cropping_flag
176 unsigned int crop_left; ///< frame_cropping_rect_left_offset
177 unsigned int crop_right; ///< frame_cropping_rect_right_offset
178 unsigned int crop_top; ///< frame_cropping_rect_top_offset
179 unsigned int crop_bottom; ///< frame_cropping_rect_bottom_offset
180 int vui_parameters_present_flag;
182 int video_signal_type_present_flag;
184 int colour_description_present_flag;
185 enum AVColorPrimaries color_primaries;
186 enum AVColorTransferCharacteristic color_trc;
187 enum AVColorSpace colorspace;
188 int timing_info_present_flag;
189 uint32_t num_units_in_tick;
191 int fixed_frame_rate_flag;
192 short offset_for_ref_frame[256]; //FIXME dyn aloc?
193 int bitstream_restriction_flag;
194 int num_reorder_frames;
195 int scaling_matrix_present;
196 uint8_t scaling_matrix4[6][16];
197 uint8_t scaling_matrix8[2][64];
198 int nal_hrd_parameters_present_flag;
199 int vcl_hrd_parameters_present_flag;
200 int pic_struct_present_flag;
201 int time_offset_length;
202 int cpb_cnt; ///< See H.264 E.1.2
203 int initial_cpb_removal_delay_length; ///< initial_cpb_removal_delay_length_minus1 +1
204 int cpb_removal_delay_length; ///< cpb_removal_delay_length_minus1 + 1
205 int dpb_output_delay_length; ///< dpb_output_delay_length_minus1 + 1
206 int bit_depth_luma; ///< bit_depth_luma_minus8 + 8
207 int bit_depth_chroma; ///< bit_depth_chroma_minus8 + 8
208 int residual_color_transform_flag; ///< residual_colour_transform_flag
212 * Picture parameter set
216 int cabac; ///< entropy_coding_mode_flag
217 int pic_order_present; ///< pic_order_present_flag
218 int slice_group_count; ///< num_slice_groups_minus1 + 1
219 int mb_slice_group_map_type;
220 unsigned int ref_count[2]; ///< num_ref_idx_l0/1_active_minus1 + 1
221 int weighted_pred; ///< weighted_pred_flag
222 int weighted_bipred_idc;
223 int init_qp; ///< pic_init_qp_minus26 + 26
224 int init_qs; ///< pic_init_qs_minus26 + 26
225 int chroma_qp_index_offset[2];
226 int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
227 int constrained_intra_pred; ///< constrained_intra_pred_flag
228 int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
229 int transform_8x8_mode; ///< transform_8x8_mode_flag
230 uint8_t scaling_matrix4[6][16];
231 uint8_t scaling_matrix8[2][64];
232 uint8_t chroma_qp_table[2][64]; ///< pre-scaled (with chroma_qp_index_offset) version of qp_table
237 * Memory management control operation opcode.
239 typedef enum MMCOOpcode{
250 * Memory management control operation.
254 int short_pic_num; ///< pic_num without wrapping (pic_num & max_pic_num)
255 int long_arg; ///< index, pic_num, or num long refs depending on opcode
261 typedef struct H264Context{
265 uint8_t *rbsp_buffer[2];
266 unsigned int rbsp_buffer_size[2];
269 * Used to parse AVC variant of h264
271 int is_avc; ///< this flag is != 0 if codec is avc1
272 int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
274 int chroma_qp[2]; //QPc
276 int qp_thresh; ///< QP threshold to skip loopfilter
282 int chroma_pred_mode;
283 int intra16x16_pred_mode;
295 const uint8_t * left_block;
296 int topleft_partition;
298 int8_t intra4x4_pred_mode_cache[5*8];
299 int8_t (*intra4x4_pred_mode)[8];
301 unsigned int topleft_samples_available;
302 unsigned int top_samples_available;
303 unsigned int topright_samples_available;
304 unsigned int left_samples_available;
305 uint8_t (*top_borders[2])[16+2*8];
306 uint8_t left_border[2*(17+2*9)];
309 * non zero coeff count cache.
310 * is 64 if not available.
312 DECLARE_ALIGNED_8(uint8_t, non_zero_count_cache)[6*8];
320 uint8_t (*non_zero_count)[32];
323 * Motion vector cache.
325 DECLARE_ALIGNED_16(int16_t, mv_cache)[2][5*8][2];
326 DECLARE_ALIGNED_8(int8_t, ref_cache)[2][5*8];
327 #define LIST_NOT_USED -1 //FIXME rename?
328 #define PART_NOT_AVAILABLE -2
331 * is 1 if the specific list MV&references are set to 0,0,-2.
333 int mv_cache_clean[2];
336 * number of neighbors (top and/or left) that used 8x8 dct
338 int neighbor_transform_size;
341 * block_offset[ 0..23] for frame macroblocks
342 * block_offset[24..47] for field macroblocks
344 int block_offset[2*(16+8)];
346 uint32_t *mb2b_xy; //FIXME are these 4 a good idea?
348 int b_stride; //FIXME use s->b4_stride
351 int mb_linesize; ///< may be equal to s->linesize or s->linesize*2, for mbaff
360 int unknown_svq3_flag;
361 int next_slice_index;
363 SPS *sps_buffers[MAX_SPS_COUNT];
364 SPS sps; ///< current sps
366 PPS *pps_buffers[MAX_PPS_COUNT];
370 PPS pps; //FIXME move to Picture perhaps? (->no) do we need that?
372 uint32_t dequant4_buffer[6][52][16];
373 uint32_t dequant8_buffer[2][52][64];
374 uint32_t (*dequant4_coeff[6])[16];
375 uint32_t (*dequant8_coeff[2])[64];
376 int dequant_coeff_pps; ///< reinit tables when pps changes
379 uint16_t *slice_table_base;
380 uint16_t *slice_table; ///< slice_table_base + 2*mb_stride + 1
382 int slice_type_nos; ///< S free slice type (SI/SP are remapped to I/P)
383 int slice_type_fixed;
385 //interlacing specific flags
387 int mb_field_decoding_flag;
388 int mb_mbaff; ///< mb_aff_frame && mb_field_decoding_flag
390 DECLARE_ALIGNED_8(uint16_t, sub_mb_type)[4];
395 int delta_poc_bottom;
398 int prev_poc_msb; ///< poc_msb of the last reference pic for POC type 0
399 int prev_poc_lsb; ///< poc_lsb of the last reference pic for POC type 0
400 int frame_num_offset; ///< for POC type 2
401 int prev_frame_num_offset; ///< for POC type 2
402 int prev_frame_num; ///< frame_num of the last pic for POC type 1/2
405 * frame_num for frames or 2*frame_num+1 for field pics.
410 * max_frame_num or 2*max_frame_num for field pics.
414 //Weighted pred stuff
416 int use_weight_chroma;
417 int luma_log2_weight_denom;
418 int chroma_log2_weight_denom;
419 int luma_weight[2][48];
420 int luma_offset[2][48];
421 int chroma_weight[2][48][2];
422 int chroma_offset[2][48][2];
423 int implicit_weight[48][48];
426 int deblocking_filter; ///< disable_deblocking_filter_idc with 1<->0
427 int slice_alpha_c0_offset;
428 int slice_beta_offset;
430 int redundant_pic_count;
432 int direct_spatial_mv_pred;
435 int dist_scale_factor[16];
436 int dist_scale_factor_field[2][32];
437 int map_col_to_list0[2][16+32];
438 int map_col_to_list0_field[2][2][16+32];
441 * num_ref_idx_l0/1_active_minus1 + 1
443 unsigned int ref_count[2]; ///< counts frames or fields, depending on current mb mode
444 unsigned int list_count;
445 uint8_t *list_counts; ///< Array of list_count per MB specifying the slice type
446 Picture *short_ref[32];
447 Picture *long_ref[32];
448 Picture default_ref_list[2][32]; ///< base reference list for all slices of a coded picture
449 Picture ref_list[2][48]; /**< 0..15: frame refs, 16..47: mbaff field refs.
450 Reordered version of default_ref_list
451 according to picture reordering in slice header */
452 int ref2frm[MAX_SLICES][2][64]; ///< reference to frame number lists, used in the loop filter, the first 2 are for -2,-1
453 Picture *delayed_pic[MAX_DELAYED_PIC_COUNT+2]; //FIXME size?
457 * memory management control operations buffer.
459 MMCO mmco[MAX_MMCO_COUNT];
462 int long_ref_count; ///< number of actual long term references
463 int short_ref_count; ///< number of actual short term references
466 GetBitContext intra_gb;
467 GetBitContext inter_gb;
468 GetBitContext *intra_gb_ptr;
469 GetBitContext *inter_gb_ptr;
471 DECLARE_ALIGNED_16(DCTELEM, mb)[16*24];
472 DCTELEM mb_padding[256]; ///< as mb is addressed by scantable[i] and scantable is uint8_t we can either check that i is not too large or ensure that there is some unused stuff after mb
478 uint8_t cabac_state[460];
481 /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
486 /* chroma_pred_mode for i4x4 or i16x16, else 0 */
487 uint8_t *chroma_pred_mode_table;
488 int last_qscale_diff;
489 uint8_t (*mvd_table[2])[2];
490 DECLARE_ALIGNED_16(uint8_t, mvd_cache)[2][5*8][2];
491 uint8_t *direct_table;
492 uint8_t direct_cache[5*8];
494 uint8_t zigzag_scan[16];
495 uint8_t zigzag_scan8x8[64];
496 uint8_t zigzag_scan8x8_cavlc[64];
497 uint8_t field_scan[16];
498 uint8_t field_scan8x8[64];
499 uint8_t field_scan8x8_cavlc[64];
500 const uint8_t *zigzag_scan_q0;
501 const uint8_t *zigzag_scan8x8_q0;
502 const uint8_t *zigzag_scan8x8_cavlc_q0;
503 const uint8_t *field_scan_q0;
504 const uint8_t *field_scan8x8_q0;
505 const uint8_t *field_scan8x8_cavlc_q0;
510 * @defgroup multithreading Members for slice based multithreading
513 struct H264Context *thread_context[MAX_THREADS];
516 * current slice number, used to initalize slice_num of each thread/context
521 * Max number of threads / contexts.
522 * This is equal to AVCodecContext.thread_count unless
523 * multithreaded decoding is impossible, in which case it is
529 * 1 if the single thread fallback warning has already been
530 * displayed, 0 otherwise.
532 int single_decode_warning;
539 uint32_t svq3_watermark_key;
542 * pic_struct in picture timing SEI message
544 SEI_PicStructType sei_pic_struct;
547 * Complement sei_pic_struct
548 * SEI_PIC_STRUCT_TOP_BOTTOM and SEI_PIC_STRUCT_BOTTOM_TOP indicate interlaced frames.
549 * However, soft telecined frames may have these values.
550 * This is used in an attempt to flag soft telecine progressive.
552 int prev_interlaced_frame;
555 * Bit set of clock types for fields/frames in picture timing SEI message.
556 * For each found ct_type, appropriate bit is set (e.g., bit 1 for
562 * dpb_output_delay in picture timing SEI message, see H.264 C.2.2
564 int sei_dpb_output_delay;
567 * cpb_removal_delay in picture timing SEI message, see H.264 C.1.2
569 int sei_cpb_removal_delay;
572 * recovery_frame_cnt from SEI message
574 * Set to -1 if no recovery point SEI message found or to number of frames
575 * before playback synchronizes. Frames having recovery point are key
578 int sei_recovery_frame_cnt;
582 int luma_weight_flag[2]; ///< 7.4.3.2 luma_weight_lX_flag
583 int chroma_weight_flag[2]; ///< 7.4.3.2 chroma_weight_lX_flag
586 int sei_buffering_period_present; ///< Buffering period SEI flag
587 int initial_cpb_removal_delay[32]; ///< Initial timestamps for CPBs
591 extern const uint8_t ff_h264_chroma_qp[52];
593 void ff_svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
595 void ff_svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
600 int ff_h264_decode_sei(H264Context *h);
605 int ff_h264_decode_seq_parameter_set(H264Context *h);
610 int ff_h264_decode_picture_parameter_set(H264Context *h, int bit_length);
613 * Decodes a network abstraction layer unit.
614 * @param consumed is the number of bytes used as input
615 * @param length is the length of the array
616 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp tailing?
617 * @returns decoded bytes, might be src+1 if no escapes
619 const uint8_t *ff_h264_decode_nal(H264Context *h, const uint8_t *src, int *dst_length, int *consumed, int length);
622 * identifies the exact end of the bitstream
623 * @return the length of the trailing, or 0 if damaged
625 int ff_h264_decode_rbsp_trailing(H264Context *h, const uint8_t *src);
628 * frees any data that may have been allocated in the H264 context like SPS, PPS etc.
630 av_cold void ff_h264_free_context(H264Context *h);
633 * reconstructs bitstream slice_type.
635 int ff_h264_get_slice_type(const H264Context *h);
641 int ff_h264_alloc_tables(H264Context *h);
644 * fills the default_ref_list.
646 int ff_h264_fill_default_ref_list(H264Context *h);
648 int ff_h264_decode_ref_pic_list_reordering(H264Context *h);
649 void ff_h264_fill_mbaff_ref_list(H264Context *h);
650 void ff_h264_remove_all_refs(H264Context *h);
653 * Executes the reference picture marking (memory management control operations).
655 int ff_h264_execute_ref_pic_marking(H264Context *h, MMCO *mmco, int mmco_count);
657 int ff_h264_decode_ref_pic_marking(H264Context *h, GetBitContext *gb);
661 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
663 int ff_h264_check_intra4x4_pred_mode(H264Context *h);
666 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
668 int ff_h264_check_intra_pred_mode(H264Context *h, int mode);
670 void ff_h264_write_back_intra_pred_mode(H264Context *h);
671 void ff_h264_hl_decode_mb(H264Context *h);
672 int ff_h264_frame_start(H264Context *h);
673 av_cold int ff_h264_decode_init(AVCodecContext *avctx);
674 av_cold int ff_h264_decode_end(AVCodecContext *avctx);
675 av_cold void ff_h264_decode_init_vlc(void);
678 * decodes a macroblock
679 * @returns 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
681 int ff_h264_decode_mb_cavlc(H264Context *h);
684 * decodes a CABAC coded macroblock
685 * @returns 0 if OK, AC_ERROR / DC_ERROR / MV_ERROR if an error is noticed
687 int ff_h264_decode_mb_cabac(H264Context *h);
689 void ff_h264_init_cabac_states(H264Context *h);
691 void ff_h264_direct_dist_scale_factor(H264Context * const h);
692 void ff_h264_direct_ref_list_init(H264Context * const h);
693 void ff_h264_pred_direct_motion(H264Context * const h, int *mb_type);
695 void ff_h264_filter_mb_fast( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize);
696 void ff_h264_filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr, unsigned int linesize, unsigned int uvlinesize);
699 * Reset SEI values at the beginning of the frame.
701 * @param h H.264 context.
703 void ff_h264_reset_sei(H264Context *h);
715 //This table must be here because scan8[constant] must be known at compiletime
716 static const uint8_t scan8[16 + 2*4]={
717 4+1*8, 5+1*8, 4+2*8, 5+2*8,
718 6+1*8, 7+1*8, 6+2*8, 7+2*8,
719 4+3*8, 5+3*8, 4+4*8, 5+4*8,
720 6+3*8, 7+3*8, 6+4*8, 7+4*8,
727 static av_always_inline uint32_t pack16to32(int a, int b){
729 return (b&0xFFFF) + (a<<16);
731 return (a&0xFFFF) + (b<<16);
735 static av_always_inline uint16_t pack8to16(int a, int b){
737 return (b&0xFF) + (a<<8);
739 return (a&0xFF) + (b<<8);
744 * gets the chroma qp.
746 static inline int get_chroma_qp(H264Context *h, int t, int qscale){
747 return h->pps.chroma_qp_table[t][qscale];
750 static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my);
752 static void fill_decode_neighbors(H264Context *h, int mb_type){
753 MpegEncContext * const s = &h->s;
754 const int mb_xy= h->mb_xy;
755 int topleft_xy, top_xy, topright_xy, left_xy[2];
756 static const uint8_t left_block_options[4][16]={
757 {0,1,2,3,7,10,8,11,7+0*8, 7+1*8, 7+2*8, 7+3*8, 2+0*8, 2+3*8, 2+1*8, 2+2*8},
758 {2,2,3,3,8,11,8,11,7+2*8, 7+2*8, 7+3*8, 7+3*8, 2+1*8, 2+2*8, 2+1*8, 2+2*8},
759 {0,0,1,1,7,10,7,10,7+0*8, 7+0*8, 7+1*8, 7+1*8, 2+0*8, 2+3*8, 2+0*8, 2+3*8},
760 {0,2,0,2,7,10,7,10,7+0*8, 7+2*8, 7+0*8, 7+2*8, 2+0*8, 2+3*8, 2+0*8, 2+3*8}
763 h->topleft_partition= -1;
765 top_xy = mb_xy - (s->mb_stride << MB_FIELD);
767 /* Wow, what a mess, why didn't they simplify the interlacing & intra
768 * stuff, I can't imagine that these complex rules are worth it. */
770 topleft_xy = top_xy - 1;
771 topright_xy= top_xy + 1;
772 left_xy[1] = left_xy[0] = mb_xy-1;
773 h->left_block = left_block_options[0];
775 const int left_mb_field_flag = IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]);
776 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
778 if (left_mb_field_flag != curr_mb_field_flag) {
779 left_xy[1] = left_xy[0] = mb_xy - s->mb_stride - 1;
780 if (curr_mb_field_flag) {
781 left_xy[1] += s->mb_stride;
782 h->left_block = left_block_options[3];
784 topleft_xy += s->mb_stride;
785 // take top left mv from the middle of the mb, as opposed to all other modes which use the bottom right partition
786 h->topleft_partition = 0;
787 h->left_block = left_block_options[1];
791 if(curr_mb_field_flag){
792 topleft_xy += s->mb_stride & (((s->current_picture.mb_type[top_xy - 1]>>7)&1)-1);
793 topright_xy += s->mb_stride & (((s->current_picture.mb_type[top_xy + 1]>>7)&1)-1);
794 top_xy += s->mb_stride & (((s->current_picture.mb_type[top_xy ]>>7)&1)-1);
796 if (left_mb_field_flag != curr_mb_field_flag) {
797 left_xy[1] = left_xy[0] = mb_xy - 1;
798 if (curr_mb_field_flag) {
799 left_xy[1] += s->mb_stride;
800 h->left_block = left_block_options[3];
802 h->left_block = left_block_options[2];
808 h->topleft_mb_xy = topleft_xy;
809 h->top_mb_xy = top_xy;
810 h->topright_mb_xy= topright_xy;
811 h->left_mb_xy[0] = left_xy[0];
812 h->left_mb_xy[1] = left_xy[1];
813 //FIXME do we need all in the context?
814 h->topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
815 h->top_type = h->slice_table[top_xy ] == h->slice_num ? s->current_picture.mb_type[top_xy] : 0;
816 h->topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
817 h->left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
818 h->left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
821 static void fill_decode_caches(H264Context *h, int mb_type){
822 MpegEncContext * const s = &h->s;
823 int topleft_xy, top_xy, topright_xy, left_xy[2];
824 int topleft_type, top_type, topright_type, left_type[2];
825 const uint8_t * left_block= h->left_block;
828 topleft_xy = h->topleft_mb_xy ;
829 top_xy = h->top_mb_xy ;
830 topright_xy = h->topright_mb_xy;
831 left_xy[0] = h->left_mb_xy[0] ;
832 left_xy[1] = h->left_mb_xy[1] ;
833 topleft_type = h->topleft_type ;
834 top_type = h->top_type ;
835 topright_type= h->topright_type ;
836 left_type[0] = h->left_type[0] ;
837 left_type[1] = h->left_type[1] ;
839 if(!IS_SKIP(mb_type)){
840 if(IS_INTRA(mb_type)){
841 int type_mask= h->pps.constrained_intra_pred ? IS_INTRA(-1) : -1;
842 h->topleft_samples_available=
843 h->top_samples_available=
844 h->left_samples_available= 0xFFFF;
845 h->topright_samples_available= 0xEEEA;
847 if(!(top_type & type_mask)){
848 h->topleft_samples_available= 0xB3FF;
849 h->top_samples_available= 0x33FF;
850 h->topright_samples_available= 0x26EA;
852 if(IS_INTERLACED(mb_type) != IS_INTERLACED(left_type[0])){
853 if(IS_INTERLACED(mb_type)){
854 if(!(left_type[0] & type_mask)){
855 h->topleft_samples_available&= 0xDFFF;
856 h->left_samples_available&= 0x5FFF;
858 if(!(left_type[1] & type_mask)){
859 h->topleft_samples_available&= 0xFF5F;
860 h->left_samples_available&= 0xFF5F;
863 int left_typei = h->slice_table[left_xy[0] + s->mb_stride ] == h->slice_num
864 ? s->current_picture.mb_type[left_xy[0] + s->mb_stride] : 0;
865 assert(left_xy[0] == left_xy[1]);
866 if(!((left_typei & type_mask) && (left_type[0] & type_mask))){
867 h->topleft_samples_available&= 0xDF5F;
868 h->left_samples_available&= 0x5F5F;
872 if(!(left_type[0] & type_mask)){
873 h->topleft_samples_available&= 0xDF5F;
874 h->left_samples_available&= 0x5F5F;
878 if(!(topleft_type & type_mask))
879 h->topleft_samples_available&= 0x7FFF;
881 if(!(topright_type & type_mask))
882 h->topright_samples_available&= 0xFBFF;
884 if(IS_INTRA4x4(mb_type)){
885 if(IS_INTRA4x4(top_type)){
886 h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
887 h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
888 h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
889 h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
892 if(!(top_type & type_mask))
897 h->intra4x4_pred_mode_cache[4+8*0]=
898 h->intra4x4_pred_mode_cache[5+8*0]=
899 h->intra4x4_pred_mode_cache[6+8*0]=
900 h->intra4x4_pred_mode_cache[7+8*0]= pred;
903 if(IS_INTRA4x4(left_type[i])){
904 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
905 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
908 if(!(left_type[i] & type_mask))
913 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
914 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
929 //FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
931 AV_COPY32(&h->non_zero_count_cache[4+8*0], &h->non_zero_count[top_xy][4+3*8]);
932 h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][1+1*8];
933 h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][2+1*8];
935 h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][1+2*8];
936 h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][2+2*8];
938 h->non_zero_count_cache[1+8*0]=
939 h->non_zero_count_cache[2+8*0]=
941 h->non_zero_count_cache[1+8*3]=
942 h->non_zero_count_cache[2+8*3]=
943 AV_WN32A(&h->non_zero_count_cache[4+8*0], CABAC && !IS_INTRA(mb_type) ? 0 : 0x40404040);
946 for (i=0; i<2; i++) {
948 h->non_zero_count_cache[3+8*1 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[8+0+2*i]];
949 h->non_zero_count_cache[3+8*2 + 2*8*i]= h->non_zero_count[left_xy[i]][left_block[8+1+2*i]];
950 h->non_zero_count_cache[0+8*1 + 8*i]= h->non_zero_count[left_xy[i]][left_block[8+4+2*i]];
951 h->non_zero_count_cache[0+8*4 + 8*i]= h->non_zero_count[left_xy[i]][left_block[8+5+2*i]];
953 h->non_zero_count_cache[3+8*1 + 2*8*i]=
954 h->non_zero_count_cache[3+8*2 + 2*8*i]=
955 h->non_zero_count_cache[0+8*1 + 8*i]=
956 h->non_zero_count_cache[0+8*4 + 8*i]= CABAC && !IS_INTRA(mb_type) ? 0 : 64;
963 h->top_cbp = h->cbp_table[top_xy];
964 } else if(IS_INTRA(mb_type)) {
971 h->left_cbp = h->cbp_table[left_xy[0]] & 0x1f0;
972 } else if(IS_INTRA(mb_type)) {
978 h->left_cbp |= ((h->cbp_table[left_xy[0]]>>((left_block[0]&(~1))+1))&0x1) << 1;
981 h->left_cbp |= ((h->cbp_table[left_xy[1]]>>((left_block[2]&(~1))+1))&0x1) << 3;
987 if(IS_INTER(mb_type) || (IS_DIRECT(mb_type) && h->direct_spatial_mv_pred)){
989 for(list=0; list<h->list_count; list++){
990 if(!USES_LIST(mb_type, list)){
991 /*if(!h->mv_cache_clean[list]){
992 memset(h->mv_cache [list], 0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
993 memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
994 h->mv_cache_clean[list]= 1;
998 assert(!(IS_DIRECT(mb_type) && !h->direct_spatial_mv_pred));
1000 h->mv_cache_clean[list]= 0;
1002 if(USES_LIST(top_type, list)){
1003 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
1004 const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
1005 AV_COPY128(h->mv_cache[list][scan8[0] + 0 - 1*8], s->current_picture.motion_val[list][b_xy + 0]);
1006 h->ref_cache[list][scan8[0] + 0 - 1*8]=
1007 h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
1008 h->ref_cache[list][scan8[0] + 2 - 1*8]=
1009 h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
1011 AV_ZERO128(h->mv_cache[list][scan8[0] + 0 - 1*8]);
1012 AV_WN32A(&h->ref_cache[list][scan8[0] + 0 - 1*8], ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101);
1016 int cache_idx = scan8[0] - 1 + i*2*8;
1017 if(USES_LIST(left_type[i], list)){
1018 const int b_xy= h->mb2b_xy[left_xy[i]] + 3;
1019 const int b8_xy= h->mb2b8_xy[left_xy[i]] + 1;
1020 AV_COPY32(h->mv_cache[list][cache_idx ], s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[0+i*2]]);
1021 AV_COPY32(h->mv_cache[list][cache_idx+8], s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[1+i*2]]);
1022 h->ref_cache[list][cache_idx ]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0+i*2]>>1)];
1023 h->ref_cache[list][cache_idx+8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[1+i*2]>>1)];
1025 AV_ZERO32(h->mv_cache [list][cache_idx ]);
1026 AV_ZERO32(h->mv_cache [list][cache_idx+8]);
1027 h->ref_cache[list][cache_idx ]=
1028 h->ref_cache[list][cache_idx+8]= (left_type[i]) ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1032 if(USES_LIST(topleft_type, list)){
1033 const int b_xy = h->mb2b_xy [topleft_xy] + 3 + h->b_stride + (h->topleft_partition & 2*h->b_stride);
1034 const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + (h->topleft_partition & h->b8_stride);
1035 AV_COPY32(h->mv_cache[list][scan8[0] - 1 - 1*8], s->current_picture.motion_val[list][b_xy]);
1036 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
1038 AV_ZERO32(h->mv_cache[list][scan8[0] - 1 - 1*8]);
1039 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1042 if(USES_LIST(topright_type, list)){
1043 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
1044 const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
1045 AV_COPY32(h->mv_cache[list][scan8[0] + 4 - 1*8], s->current_picture.motion_val[list][b_xy]);
1046 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
1048 AV_ZERO32(h->mv_cache [list][scan8[0] + 4 - 1*8]);
1049 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
1052 if((mb_type&(MB_TYPE_SKIP|MB_TYPE_DIRECT2)) && !FRAME_MBAFF)
1055 if(!(mb_type&(MB_TYPE_SKIP|MB_TYPE_DIRECT2))) {
1056 h->ref_cache[list][scan8[5 ]+1] =
1057 h->ref_cache[list][scan8[7 ]+1] =
1058 h->ref_cache[list][scan8[13]+1] = //FIXME remove past 3 (init somewhere else)
1059 h->ref_cache[list][scan8[4 ]] =
1060 h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
1061 AV_ZERO32(h->mv_cache [list][scan8[5 ]+1]);
1062 AV_ZERO32(h->mv_cache [list][scan8[7 ]+1]);
1063 AV_ZERO32(h->mv_cache [list][scan8[13]+1]); //FIXME remove past 3 (init somewhere else)
1064 AV_ZERO32(h->mv_cache [list][scan8[4 ]]);
1065 AV_ZERO32(h->mv_cache [list][scan8[12]]);
1068 /* XXX beurk, Load mvd */
1069 if(USES_LIST(top_type, list)){
1070 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
1071 AV_COPY64(h->mvd_cache[list][scan8[0] + 0 - 1*8], h->mvd_table[list][b_xy + 0]);
1073 AV_ZERO64(h->mvd_cache[list][scan8[0] + 0 - 1*8]);
1075 if(USES_LIST(left_type[0], list)){
1076 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
1077 AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 0*8], h->mvd_table[list][b_xy + h->b_stride*left_block[0]]);
1078 AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 1*8], h->mvd_table[list][b_xy + h->b_stride*left_block[1]]);
1080 AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 0*8]);
1081 AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 1*8]);
1083 if(USES_LIST(left_type[1], list)){
1084 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
1085 AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 2*8], h->mvd_table[list][b_xy + h->b_stride*left_block[2]]);
1086 AV_COPY16(h->mvd_cache[list][scan8[0] - 1 + 3*8], h->mvd_table[list][b_xy + h->b_stride*left_block[3]]);
1088 AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 2*8]);
1089 AV_ZERO16(h->mvd_cache [list][scan8[0] - 1 + 3*8]);
1091 AV_ZERO16(h->mvd_cache [list][scan8[5 ]+1]);
1092 AV_ZERO16(h->mvd_cache [list][scan8[7 ]+1]);
1093 AV_ZERO16(h->mvd_cache [list][scan8[13]+1]); //FIXME remove past 3 (init somewhere else)
1094 AV_ZERO16(h->mvd_cache [list][scan8[4 ]]);
1095 AV_ZERO16(h->mvd_cache [list][scan8[12]]);
1096 if(h->slice_type_nos == FF_B_TYPE){
1097 fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, MB_TYPE_16x16>>1, 1);
1099 if(IS_DIRECT(top_type)){
1100 AV_WN32A(&h->direct_cache[scan8[0] - 1*8], 0x01010101*(MB_TYPE_DIRECT2>>1));
1101 }else if(IS_8X8(top_type)){
1102 int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
1103 h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
1104 h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
1106 AV_WN32A(&h->direct_cache[scan8[0] - 1*8], 0x01010101*(MB_TYPE_16x16>>1));
1109 if(IS_DIRECT(left_type[0]))
1110 h->direct_cache[scan8[0] - 1 + 0*8]= MB_TYPE_DIRECT2>>1;
1111 else if(IS_8X8(left_type[0]))
1112 h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[h->mb2b8_xy[left_xy[0]] + 1 + h->b8_stride*(left_block[0]>>1)];
1114 h->direct_cache[scan8[0] - 1 + 0*8]= MB_TYPE_16x16>>1;
1116 if(IS_DIRECT(left_type[1]))
1117 h->direct_cache[scan8[0] - 1 + 2*8]= MB_TYPE_DIRECT2>>1;
1118 else if(IS_8X8(left_type[1]))
1119 h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[h->mb2b8_xy[left_xy[1]] + 1 + h->b8_stride*(left_block[2]>>1)];
1121 h->direct_cache[scan8[0] - 1 + 2*8]= MB_TYPE_16x16>>1;
1127 MAP_F2F(scan8[0] - 1 - 1*8, topleft_type)\
1128 MAP_F2F(scan8[0] + 0 - 1*8, top_type)\
1129 MAP_F2F(scan8[0] + 1 - 1*8, top_type)\
1130 MAP_F2F(scan8[0] + 2 - 1*8, top_type)\
1131 MAP_F2F(scan8[0] + 3 - 1*8, top_type)\
1132 MAP_F2F(scan8[0] + 4 - 1*8, topright_type)\
1133 MAP_F2F(scan8[0] - 1 + 0*8, left_type[0])\
1134 MAP_F2F(scan8[0] - 1 + 1*8, left_type[0])\
1135 MAP_F2F(scan8[0] - 1 + 2*8, left_type[1])\
1136 MAP_F2F(scan8[0] - 1 + 3*8, left_type[1])
1138 #define MAP_F2F(idx, mb_type)\
1139 if(!IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
1140 h->ref_cache[list][idx] <<= 1;\
1141 h->mv_cache[list][idx][1] /= 2;\
1142 h->mvd_cache[list][idx][1] >>=1;\
1147 #define MAP_F2F(idx, mb_type)\
1148 if(IS_INTERLACED(mb_type) && h->ref_cache[list][idx] >= 0){\
1149 h->ref_cache[list][idx] >>= 1;\
1150 h->mv_cache[list][idx][1] <<= 1;\
1151 h->mvd_cache[list][idx][1] <<= 1;\
1161 h->neighbor_transform_size= !!IS_8x8DCT(top_type) + !!IS_8x8DCT(left_type[0]);
1166 * @returns non zero if the loop filter can be skiped
1168 static int fill_filter_caches(H264Context *h, int mb_type){
1169 MpegEncContext * const s = &h->s;
1170 const int mb_xy= h->mb_xy;
1171 int top_xy, left_xy[2];
1172 int top_type, left_type[2];
1174 top_xy = mb_xy - (s->mb_stride << MB_FIELD);
1176 //FIXME deblocking could skip the intra and nnz parts.
1178 /* Wow, what a mess, why didn't they simplify the interlacing & intra
1179 * stuff, I can't imagine that these complex rules are worth it. */
1181 left_xy[1] = left_xy[0] = mb_xy-1;
1183 const int left_mb_field_flag = IS_INTERLACED(s->current_picture.mb_type[mb_xy-1]);
1184 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
1186 if (left_mb_field_flag != curr_mb_field_flag) {
1187 left_xy[0] -= s->mb_stride;
1190 if(curr_mb_field_flag){
1191 top_xy += s->mb_stride & (((s->current_picture.mb_type[top_xy ]>>7)&1)-1);
1193 if (left_mb_field_flag != curr_mb_field_flag) {
1194 left_xy[1] += s->mb_stride;
1199 h->top_mb_xy = top_xy;
1200 h->left_mb_xy[0] = left_xy[0];
1201 h->left_mb_xy[1] = left_xy[1];
1203 //for sufficiently low qp, filtering wouldn't do anything
1204 //this is a conservative estimate: could also check beta_offset and more accurate chroma_qp
1205 int qp_thresh = h->qp_thresh; //FIXME strictly we should store qp_thresh for each mb of a slice
1206 int qp = s->current_picture.qscale_table[mb_xy];
1208 && (left_xy[0]<0 || ((qp + s->current_picture.qscale_table[left_xy[0]] + 1)>>1) <= qp_thresh)
1209 && (top_xy < 0 || ((qp + s->current_picture.qscale_table[top_xy ] + 1)>>1) <= qp_thresh)){
1212 if( (left_xy[0]< 0 || ((qp + s->current_picture.qscale_table[left_xy[1] ] + 1)>>1) <= qp_thresh)
1213 && (top_xy < s->mb_stride || ((qp + s->current_picture.qscale_table[top_xy -s->mb_stride] + 1)>>1) <= qp_thresh))
1218 if(h->deblocking_filter == 2){
1219 h->top_type = top_type = h->slice_table[top_xy ] == h->slice_num ? s->current_picture.mb_type[top_xy] : 0;
1220 h->left_type[0]= left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
1221 h->left_type[1]= left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
1223 h->top_type = top_type = h->slice_table[top_xy ] < 0xFFFF ? s->current_picture.mb_type[top_xy] : 0;
1224 h->left_type[0]= left_type[0] = h->slice_table[left_xy[0] ] < 0xFFFF ? s->current_picture.mb_type[left_xy[0]] : 0;
1225 h->left_type[1]= left_type[1] = h->slice_table[left_xy[1] ] < 0xFFFF ? s->current_picture.mb_type[left_xy[1]] : 0;
1227 if(IS_INTRA(mb_type))
1230 AV_COPY64(&h->non_zero_count_cache[0+8*1], &h->non_zero_count[mb_xy][ 0]);
1231 AV_COPY64(&h->non_zero_count_cache[0+8*2], &h->non_zero_count[mb_xy][ 8]);
1232 AV_COPY32(&h->non_zero_count_cache[0+8*5], &h->non_zero_count[mb_xy][16]);
1233 AV_COPY32(&h->non_zero_count_cache[4+8*3], &h->non_zero_count[mb_xy][20]);
1234 AV_COPY64(&h->non_zero_count_cache[0+8*4], &h->non_zero_count[mb_xy][24]);
1236 h->cbp= h->cbp_table[mb_xy];
1240 for(list=0; list<h->list_count; list++){
1243 int16_t (*mv_dst)[2];
1244 int16_t (*mv_src)[2];
1246 if(!USES_LIST(mb_type, list)){
1247 fill_rectangle( h->mv_cache[list][scan8[0]], 4, 4, 8, pack16to32(0,0), 4);
1248 AV_WN32A(&h->ref_cache[list][scan8[ 0]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
1249 AV_WN32A(&h->ref_cache[list][scan8[ 2]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
1250 AV_WN32A(&h->ref_cache[list][scan8[ 8]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
1251 AV_WN32A(&h->ref_cache[list][scan8[10]], ((LIST_NOT_USED)&0xFF)*0x01010101u);
1255 ref = &s->current_picture.ref_index[list][h->mb2b8_xy[mb_xy]];
1257 int (*ref2frm)[64] = h->ref2frm[ h->slice_num&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
1258 AV_WN32A(&h->ref_cache[list][scan8[ 0]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
1259 AV_WN32A(&h->ref_cache[list][scan8[ 2]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
1260 ref += h->b8_stride;
1261 AV_WN32A(&h->ref_cache[list][scan8[ 8]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
1262 AV_WN32A(&h->ref_cache[list][scan8[10]], (pack16to32(ref2frm[list][ref[0]],ref2frm[list][ref[1]])&0x00FF00FF)*0x0101);
1265 b_stride = h->b_stride;
1266 mv_dst = &h->mv_cache[list][scan8[0]];
1267 mv_src = &s->current_picture.motion_val[list][4*s->mb_x + 4*s->mb_y*b_stride];
1269 AV_COPY128(mv_dst + 8*y, mv_src + y*b_stride);
1284 //FIXME constraint_intra_pred & partitioning & nnz (let us hope this is just a typo in the spec)
1286 AV_COPY32(&h->non_zero_count_cache[4+8*0], &h->non_zero_count[top_xy][4+3*8]);
1290 h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][7+0*8];
1291 h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][7+1*8];
1292 h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[0]][7+2*8];
1293 h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[0]][7+3*8];
1296 // CAVLC 8x8dct requires NNZ values for residual decoding that differ from what the loop filter needs
1297 if(!CABAC && h->pps.transform_8x8_mode){
1298 if(IS_8x8DCT(top_type)){
1299 h->non_zero_count_cache[4+8*0]=
1300 h->non_zero_count_cache[5+8*0]= h->cbp_table[top_xy] & 4;
1301 h->non_zero_count_cache[6+8*0]=
1302 h->non_zero_count_cache[7+8*0]= h->cbp_table[top_xy] & 8;
1304 if(IS_8x8DCT(left_type[0])){
1305 h->non_zero_count_cache[3+8*1]=
1306 h->non_zero_count_cache[3+8*2]= h->cbp_table[left_xy[0]]&2; //FIXME check MBAFF
1308 if(IS_8x8DCT(left_type[1])){
1309 h->non_zero_count_cache[3+8*3]=
1310 h->non_zero_count_cache[3+8*4]= h->cbp_table[left_xy[1]]&8; //FIXME check MBAFF
1313 if(IS_8x8DCT(mb_type)){
1314 h->non_zero_count_cache[scan8[0 ]]= h->non_zero_count_cache[scan8[1 ]]=
1315 h->non_zero_count_cache[scan8[2 ]]= h->non_zero_count_cache[scan8[3 ]]= h->cbp & 1;
1317 h->non_zero_count_cache[scan8[0+ 4]]= h->non_zero_count_cache[scan8[1+ 4]]=
1318 h->non_zero_count_cache[scan8[2+ 4]]= h->non_zero_count_cache[scan8[3+ 4]]= h->cbp & 2;
1320 h->non_zero_count_cache[scan8[0+ 8]]= h->non_zero_count_cache[scan8[1+ 8]]=
1321 h->non_zero_count_cache[scan8[2+ 8]]= h->non_zero_count_cache[scan8[3+ 8]]= h->cbp & 4;
1323 h->non_zero_count_cache[scan8[0+12]]= h->non_zero_count_cache[scan8[1+12]]=
1324 h->non_zero_count_cache[scan8[2+12]]= h->non_zero_count_cache[scan8[3+12]]= h->cbp & 8;
1328 if(IS_INTER(mb_type) || IS_DIRECT(mb_type)){
1330 for(list=0; list<h->list_count; list++){
1331 if(USES_LIST(top_type, list)){
1332 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
1333 const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
1334 int (*ref2frm)[64] = h->ref2frm[ h->slice_table[top_xy]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
1335 AV_COPY128(h->mv_cache[list][scan8[0] + 0 - 1*8], s->current_picture.motion_val[list][b_xy + 0]);
1336 h->ref_cache[list][scan8[0] + 0 - 1*8]=
1337 h->ref_cache[list][scan8[0] + 1 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 0]];
1338 h->ref_cache[list][scan8[0] + 2 - 1*8]=
1339 h->ref_cache[list][scan8[0] + 3 - 1*8]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + 1]];
1341 AV_ZERO128(h->mv_cache[list][scan8[0] + 0 - 1*8]);
1342 AV_WN32A(&h->ref_cache[list][scan8[0] + 0 - 1*8], ((LIST_NOT_USED)&0xFF)*0x01010101u);
1345 if(!IS_INTERLACED(mb_type^left_type[0])){
1346 if(USES_LIST(left_type[0], list)){
1347 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
1348 const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;
1349 int (*ref2frm)[64] = h->ref2frm[ h->slice_table[left_xy[0]]&(MAX_SLICES-1) ][0] + (MB_MBAFF ? 20 : 2);
1350 AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 0 ], s->current_picture.motion_val[list][b_xy + h->b_stride*0]);
1351 AV_COPY32(h->mv_cache[list][scan8[0] - 1 + 8 ], s->current_picture.motion_val[list][b_xy + h->b_stride*1]);
1352 AV_COPY32(h->mv_cache[list][scan8[0] - 1 +16 ], s->current_picture.motion_val[list][b_xy + h->b_stride*2]);
1353 AV_COPY32(h->mv_cache[list][scan8[0] - 1 +24 ], s->current_picture.motion_val[list][b_xy + h->b_stride*3]);
1354 h->ref_cache[list][scan8[0] - 1 + 0 ]=
1355 h->ref_cache[list][scan8[0] - 1 + 8 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + h->b8_stride*0]];
1356 h->ref_cache[list][scan8[0] - 1 +16 ]=
1357 h->ref_cache[list][scan8[0] - 1 +24 ]= ref2frm[list][s->current_picture.ref_index[list][b8_xy + h->b8_stride*1]];
1359 AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 0 ]);
1360 AV_ZERO32(h->mv_cache [list][scan8[0] - 1 + 8 ]);
1361 AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +16 ]);
1362 AV_ZERO32(h->mv_cache [list][scan8[0] - 1 +24 ]);
1363 h->ref_cache[list][scan8[0] - 1 + 0 ]=
1364 h->ref_cache[list][scan8[0] - 1 + 8 ]=
1365 h->ref_cache[list][scan8[0] - 1 + 16 ]=
1366 h->ref_cache[list][scan8[0] - 1 + 24 ]= LIST_NOT_USED;
1376 * gets the predicted intra4x4 prediction mode.
1378 static inline int pred_intra_mode(H264Context *h, int n){
1379 const int index8= scan8[n];
1380 const int left= h->intra4x4_pred_mode_cache[index8 - 1];
1381 const int top = h->intra4x4_pred_mode_cache[index8 - 8];
1382 const int min= FFMIN(left, top);
1384 tprintf(h->s.avctx, "mode:%d %d min:%d\n", left ,top, min);
1386 if(min<0) return DC_PRED;
1390 static inline void write_back_non_zero_count(H264Context *h){
1391 const int mb_xy= h->mb_xy;
1393 AV_COPY64(&h->non_zero_count[mb_xy][ 0], &h->non_zero_count_cache[0+8*1]);
1394 AV_COPY64(&h->non_zero_count[mb_xy][ 8], &h->non_zero_count_cache[0+8*2]);
1395 AV_COPY32(&h->non_zero_count[mb_xy][16], &h->non_zero_count_cache[0+8*5]);
1396 AV_COPY32(&h->non_zero_count[mb_xy][20], &h->non_zero_count_cache[4+8*3]);
1397 AV_COPY64(&h->non_zero_count[mb_xy][24], &h->non_zero_count_cache[0+8*4]);
1400 static inline void write_back_motion(H264Context *h, int mb_type){
1401 MpegEncContext * const s = &h->s;
1402 const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1403 const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1406 if(!USES_LIST(mb_type, 0))
1407 fill_rectangle(&s->current_picture.ref_index[0][b8_xy], 2, 2, h->b8_stride, (uint8_t)LIST_NOT_USED, 1);
1409 for(list=0; list<h->list_count; list++){
1411 int16_t (*mv_dst)[2];
1412 int16_t (*mv_src)[2];
1414 if(!USES_LIST(mb_type, list))
1417 b_stride = h->b_stride;
1418 mv_dst = &s->current_picture.motion_val[list][b_xy];
1419 mv_src = &h->mv_cache[list][scan8[0]];
1421 AV_COPY128(mv_dst + y*b_stride, mv_src + 8*y);
1424 uint8_t (*mvd_dst)[2] = &h->mvd_table[list][b_xy];
1425 uint8_t (*mvd_src)[2] = &h->mvd_cache[list][scan8[0]];
1426 if(IS_SKIP(mb_type))
1427 fill_rectangle(mvd_dst, 4, 4, h->b_stride, 0, 2);
1430 AV_COPY64(mvd_dst + y*b_stride, mvd_src + 8*y);
1435 int8_t *ref_index = &s->current_picture.ref_index[list][b8_xy];
1436 ref_index[0+0*h->b8_stride]= h->ref_cache[list][scan8[0]];
1437 ref_index[1+0*h->b8_stride]= h->ref_cache[list][scan8[4]];
1438 ref_index[0+1*h->b8_stride]= h->ref_cache[list][scan8[8]];
1439 ref_index[1+1*h->b8_stride]= h->ref_cache[list][scan8[12]];
1443 if(h->slice_type_nos == FF_B_TYPE && CABAC){
1444 if(IS_8X8(mb_type)){
1445 uint8_t *direct_table = &h->direct_table[b8_xy];
1446 direct_table[1+0*h->b8_stride] = h->sub_mb_type[1]>>1;
1447 direct_table[0+1*h->b8_stride] = h->sub_mb_type[2]>>1;
1448 direct_table[1+1*h->b8_stride] = h->sub_mb_type[3]>>1;
1453 static inline int get_dct8x8_allowed(H264Context *h){
1454 if(h->sps.direct_8x8_inference_flag)
1455 return !(AV_RN64A(h->sub_mb_type) & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8 )*0x0001000100010001ULL));
1457 return !(AV_RN64A(h->sub_mb_type) & ((MB_TYPE_16x8|MB_TYPE_8x16|MB_TYPE_8x8|MB_TYPE_DIRECT2)*0x0001000100010001ULL));
1461 * decodes a P_SKIP or B_SKIP macroblock
1463 static void decode_mb_skip(H264Context *h){
1464 MpegEncContext * const s = &h->s;
1465 const int mb_xy= h->mb_xy;
1468 memset(h->non_zero_count[mb_xy], 0, 32);
1469 memset(h->non_zero_count_cache + 8, 0, 8*5); //FIXME ugly, remove pfui
1472 mb_type|= MB_TYPE_INTERLACED;
1474 if( h->slice_type_nos == FF_B_TYPE )
1476 // just for fill_caches. pred_direct_motion will set the real mb_type
1477 mb_type|= MB_TYPE_L0L1|MB_TYPE_DIRECT2|MB_TYPE_SKIP;
1478 if(h->direct_spatial_mv_pred){
1479 fill_decode_neighbors(h, mb_type);
1480 fill_decode_caches(h, mb_type); //FIXME check what is needed and what not ...
1482 ff_h264_pred_direct_motion(h, &mb_type);
1483 mb_type|= MB_TYPE_SKIP;
1488 mb_type|= MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P1L0|MB_TYPE_SKIP;
1490 fill_decode_neighbors(h, mb_type);
1491 fill_decode_caches(h, mb_type); //FIXME check what is needed and what not ...
1492 pred_pskip_motion(h, &mx, &my);
1493 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
1494 fill_rectangle( h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mx,my), 4);
1497 write_back_motion(h, mb_type);
1498 s->current_picture.mb_type[mb_xy]= mb_type;
1499 s->current_picture.qscale_table[mb_xy]= s->qscale;
1500 h->slice_table[ mb_xy ]= h->slice_num;
1501 h->prev_mb_skipped= 1;
1504 #include "h264_mvpred.h" //For pred_pskip_motion()
1506 #endif /* AVCODEC_H264_H */